Structure and Function of ArnD. A Deformylase Essential for Lipid A Modification with 4-Amino-4-deoxy-L-arabinose and Polymyxin Resistance

Abstract

Covalent modification of lipid A with 4-deoxy-4-amino-L-arabinose (Ara4N) mediates resistance to cationic antimicrobial peptides and polymyxin antibiotics in Gram-negative bacteria. The proteins required for Ara4N biosynthesis are encoded in the pmrE and arnBCADTEF loci, with ArnT ultimately transferring the amino sugar from undecaprenyl-phospho-4-deoxy-4-amino-L-arabinose (C55P-Ara4N) to lipid A. However, Ara4N is N-formylated prior to its transfer to undecaprenyl-phosphate by ArnC, requiring a deformylase activity downstream in the pathway to generate the final C55P-Ara4N donor. Here, we show that deletion of the arnD gene in an Escherichia coli mutant that constitutively expresses the arnBCADTEF operon leads to accumulation of the formylated ArnC product, undecaprenyl-phospho-4-deoxy-4-formamido-L-arabinose (C55P-Ara4FN), suggesting that ArnD is the downstream deformylase. Purification of Salmonella Typhimurium ArnD (stArnD) shows it is membrane associated. We present the crystal structure of stArnD revealing a NodB homology domain structure characteristic of the metal-dependent carbohydrate esterase family 4 (CE4). However, ArnD displays several distinct features: a 44 amino acid insertion, a C-terminal extension in the NodB fold, and sequence divergence in the five motifs that define the CE4 family, suggesting that ArnD represents a new family of carbohydrate esterases. The insertion is responsible for membrane association as its deletion results in a soluble ArnD variant. The active site retains a metal coordination H-H-D triad and, in the presence of Co²⁺ or Mn²⁺, purified stArnD efficiently deformylates C55P-Ara4FN confirming its role in Ara4N biosynthesis. Mutations D9N and H233Y completely inactivate stArnD implicating these two residues in a metal-assisted acid-base catalytic mechanism.

Introduction

Lipid A is the conserved, membrane embedded component of lipopolysaccharide (LPS) that constitutes the outer leaflet of the outer membrane in diderm bacteria. Cationic antimicrobial peptides of the innate immune system as well as related antibiotics such as polymyxin B and colistin (polymyxin E), electrostatically interact with negatively charged phosphate groups in lipid A as the initial step of their antimicrobial mechanism. However, environmental conditions, such as low pH or low Mg²⁺, or bacterial exposure to cationic antimicrobial peptides, can trigger lipid A modification with cationic moieties that reduce antimicrobial binding leading to resistance^{1, 2}. One of the most common modifications is the addition of 4-amino-4-deoxy-L-arabinose (Ara4N) to phosphate groups in lipid A (SI Fig. S1), which reduces its net charge from −1.5 to 0^1–3. Ara4N can be present in more than 85% of LPS molecules⁴ and the less negatively charged cell surface impairs binding of CAMPs causing resistance^{5, 6}. The second most common modification is the addition of phosphoethanolamine (pEtN) to phosphate groups of lipid A (SI Fig. S1). As this reduces lipid A net charge from −1.5 to −1, it is less effective than Ara4N at reducing colistin binding^{2, 3}.

Addition of Ara4N to lipid A is controlled by the PmrA/PmrB and PhoP/PhoQ two-component regulatory systems^{7, 8}. Activation of PmrA results in transcription of pmrE(ugd) and the arnBCADTEF operon (previously known as pmrHFIJKLM) encoding the enzymes responsible for lipid A modification with Ara4N and polymyxin resistance^9–11. Constitutively active pmrA (pmrA^c) mutants of E. coli and Salmonella Typhimurium are thus polymyxin resistant^{4, 12, 13}. All the gene products of pmrE and the arnBCADTEF are essential for resistance as non-polar inactivation of any one of those genes results in loss of Ara4N-lipid A modification and polymyxin resistance in pmrAc cells^{9–11, 14}.

The pathway for synthesis of Ara4N and its attachment to lipid A begins with UDP-Glucose and its conversion to UDP-Ara4N by the successive action of Ugd (PmrE), the C-terminal domain of ArnA and ArnB (SI Fig. S2). The nucleotide-linked amino sugar is then formylated at the 4’ nitrogen to yield UDP-L-AraN-formyl (UDP-Ara4FN) in a reaction catalyzed by the N-terminal domain of ArnA. As no Ara4FN has been detected attached to lipid A, the reason for an apparently transient N-formylation was puzzling. However, Breazeale and coworkers have suggested that formylation of UDP-Ara4N compensates for the unfavorable equilibrium constant observed for the conversion of UDP-4-ketopentose to UDP-Ara4N, thereby displacing the reaction in the forward direction of the pathway¹⁵. ArnC then transfers the formylated amino-sugar to undecaprenyl-phosphate (C55P) to yield undecaprenyl-phospho-4-deoxy-4-formamido-L-arabinose (C55P-Ara4FN) with release of UDP¹⁶. Based on weak sequence similarity with polysaccharide deacetylases, the gene product of arnD was proposed to deformylate the intermediate C55P-Ara4FN¹⁶, and ArnD from Burkholderia cenocepacia has been reported to slowly deformylate chemically synthesized neryl-phosphate-L-Ara4FN (C10P-Ara4FN), only after a 20 hour incubation¹⁷. Deformylation of C55P-Ara4FN to yield undecaprenyl-phospho-4-deoxy-4-amino-L-arabinose (C55P-Ara4N) is essential for the lipid A modification as this intermediate is flipped to the periplasmic side of the inner membrane by the heterodimeric ArnE/ArnF^{14, 16} before transfer of the Ara4N to lipid A is catalyzed by the transmembrane protein ArnT^18–20.

Here, we show that polymyxin resistant E. coli pmrA^c accumulates the amino sugar donor C55P-Ara4N required for lipid A modification with Ara4N. However, deletion of arnD (pmrA^c ΔarnD) results in accumulation of the formylated intermediate C55P-Ara4FN indicating that ArnD is responsible for deformylating the ArnC product in cells. We present the crystal structure of ArnD from Salmonella Typhimurium and identify conserved structural motifs that define these proteins as a new family of membrane associated carbohydrate esterases. Structure analysis also reveals a conserved membrane interaction finger in ArnD that mediates the peripheral membrane association of the protein. Reconstitution of the purified ArnD activity in vitro shows efficient deformylation of C55P-[¹⁴C]Ara4FN as well as the ten carbon diprenol derivative neryl-P-L-[¹⁴C]Ara4FN in the presence of Co²⁺ or Mn²⁺. Furthermore, mutation of conserved aspartate and histidine amino acids in the active site render the enzyme inactive implicating these residues in a metal-assisted, acid-based catalyzed reaction mechanism. The unambiguous demonstration of the role of ArnD in lipid A modification with Ara4N and polymyxin resistance as well as the structural and in vitro functional characterization of the enzyme, will facilitate specific inhibition studies that may be used to combat polymyxin resistance in Gram-negative pathogens.

Materials and Methods

Construction of a pmrA^c mutant of E. coli K-12 with chromosomal insertions in arn genes.

Using a pmrA^c variant of E. coli strain DY-330 designated MST100¹⁶, we took advantage of λ-RED-mediated recombination, to make a non-polar insertion of a kanamycin resistance gene (kan, aminoglycoside 3’-phosphotransferase) into the arnD gene located in the arnBCADTEF operon. A linear DNA fragment in which the kan gene is flanked by 44 bp located directly upstream of arnD (underlined sequence in DKanKO5) and a ribosomal binding site with an NdeI site (double underlined in DKanKO3) site, followed by 45 bp located directly downstream of arnD (underlined in DKanKO3) was amplified using the oligonucleotides DKanKO3 (5’-gcaggcaataaacgcgaagaggccgataaggtaacgtaccgatttcatatgtatatctccttcttagaaaaactcatcgagcat-3’), DKanKOK5 (5’–ctggatttcttcctgcgcaccgttgatcttacggataaaccatcatgaccaaaattcaacgggaaacgtcttgc- 3’) and the kanamycin resistance gene (Tn903) from plasmid pWSK130 as a template in a PCR reaction using Pfu polymerase. The PCR product was gel purified and 100 ng were introduced into strain MST100 by electroporation following induction of the RED system by incubation of mid-log-phase cells for 15 min at 42 °C. After growth for 2 h at 30 °C, the cells were plated on LB agar containing kanamycin grown overnight at 30 °C. The resulting strain pmrA^c ΔarnD has the phenotype Pm^S, Kan^R, and only grows at 30 °C. The same approach was utilized to construct the pmrA^c ΔarnC strain.

Lipid Extraction from E. coli Strains

The lipid extraction was performed using the method of Bligh-Dyer, essentially as previously described¹³. Briefly, 400-mL cultures of bacteria were grown to A₆₀₀ = 2.0 and harvested by centrifugation at 6500 × g for 15 minutes at 4 °C. Cell pellets were then washed with phosphate-buffered saline, pH 7.4 (50 mL). To extract the phospholipids, the cell pellet was resuspended in 19 mL of a single-phase Bligh/Dyer mixture, consisting of chloroform/methanol/water (1:2:0.8 v/v). After incubating for 60 minutes at room temperature, with mixing, the insoluble material was removed by centrifugation (4000 × g for 15 minutes at room temperature), and the lipids-containing supernatant was transferred to a clean tube. The supernatant was then converted to a two-phase Bligh/Dyer system consisting of chloroform/methanol/water (2:2:1.8, v/v). The phases were separated by low-speed centrifugation (4000 × g for 15 minutes at room temperature) and the lower phase was removed, and ultimately dried under a stream of N2.

Lipid analysis by electrospray ionization/mass spectrometry (ESI/MS)

All mass spectra were acquired on a QSTAR XL quadrupole time-of-flight tandem mass spectrometer (ABI/MDS-Sciex, Toronto, Canada), equipped with an electrospray ionization source. Spectra were acquired in the negative ion mode and were typically the accumulation of 60 scans, collected from 200 to 2000 atomic mass units. Lipid extracts from the E. coli strains were resuspended in 100 μL of CHCl₃/MeOH (2:1, v/v) and 20 μL of this solution were diluted with 180 μL of CHCl₃/MeOH (1:1, v/v) and infused into the ion source at 5–10 μl/min. For the MST100 pmrA^c strain sample, 2 μL of piperidine were added to the solution to enhance the production of negatively charged, deprotonated [M-H]⁻ ion signals from neutral species. The negative ion analysis was carried out at −4200 V. Data acquisition and analysis were performed using Analyst QS software.

Cloning and expression and purification of stArnD.

The gene for Salmonella Typhimurium ArnD (stArnD, Uniprot: O52326) was PCR amplified from genomic DNA with a sense primer incorporating an NdeI restriction site at the start ATG codon, and an antisense primer coding for an hexahistidine tag immediately after the last codon followed by a stop codon and an XmaI restriction site. The PCR product was purified, digested with NdeI and XmaI (New England BioLabs) and ligated into an engineered pET28 vector digested with the same enzymes, resulting in the ArnD expression plasmid pMS1337. E. coli BL21 (DE3) cells (Invitrogen) were transformed with pMS1337, and a single colony used to inoculate 100 mL of Luria Broth (LB) media supplemented with 50 μg/mL kanamycin. The culture was incubated overnight at 37 °C and used to inoculate 4 x 1 L of LB media supplemented with 50 μg/mL kanamycin. The cells were grown at 37°C until the OD₆₀₀ = 0.6. The cultures were then cooled down on ice before adding IPTG (final concentration 0.5 mM) to induce protein expression. The cultures were then incubated overnight at room temperature with shaking. Cells were harvested by centrifugation at 4000 rpm (JLA 8.1000 rotor) for 30 minutes at 4°C. Cell pellets were resuspended in 50 mL of lysis buffer containing 25 mM Tris-HCl pH 8.0, 300 mM NaCl, 10% (v/v) glycerol, 5 mM β-mercaptoethanol (BME), lysozyme 0.5 mg/mL and 1 EDTA-free protease inhibitor cocktail tablet (Roche). The cells were lysed by 3 to 4 passes through an Avestin Emulsiflex C3 unit followed by removal of cell debris by centrifugation at 15000 rpm (Sorvall SS-34 rotor) for 30 minutes at 4°C. The supernatant was ultracentrifuged at 50000 rpm (Beckman fixed-angle Titanium rotor Type 70) for 2 hours at 4°C. The supernatant was discarded, and the membrane pellet resuspended in lysis buffer without lysozyme, supplemented with n-Dodecyl-β-D-Maltopyranoside (DDM, Anatrace) to a final concentration of 1% (w/v) and homogenized in a Dounce homogenizer. The homogenate was incubated overnight at 4°C with gentle rocking to extract ArnD. The suspension was then ultracentrifuged at 50000 rpm (Type-70Ti) for 1 hour at 4°C and the insoluble pellet discarded. The supernatant was loaded to a 5 mL HisTrap FF Ni-NTA column (Cytiva) equilibrated in 25 mM Tris-Base pH 8.0, 300 mM NaCl, 10% (v/v) glycerol, 5 mM BME and 0.023% (w/v) DDM. The column was washed with 5 column volumes of the above buffer supplemented with 5 mM imidazole and the elution was achieved by increasing the imidazole concentration to 300 mM during a five-column volume linear gradient. Fractions containing ArnD, as evaluated by SDS-PAGE, were pooled, loaded in a HiLoad 26/60 Superdex 200 (Amersham Pharmacia Biotech) column equilibrated in 25 mM Tris-Base pH 8.0, 150 mM NaCl, 10% (v/v) glycerol, 5 mM BME, 1 mM EDTA and 0.02% (w/v) DDM and eluted in the same buffer. Fractions containing ArnD, as detected by SDS-PAGE, were pooled and concentrated to 13 mg/mL using a 100k MWCO centrifugal filter unit. Protein stocks were snap-frozen in liquid nitrogen and kept at −80°C for further use.

Cloning, expression, and purification of ArnD mutants.

Point mutations D9N and H233Y in the gene for stArnD were introduced in the pMS 1337 plasmid using a variation of the QuikChange protocol²¹ and primers 5’-cgcattaatgttgataccttgcgaggaacg-3’ and 5’-atcaacattaatgcgtaaaccgactttcgtcatatg-3’ for D9N, and 5’-gtatataccatctacgcggaagtagaaggtattgtccatcag-3’ and 5’-cgcgtagatggtatataccggcgtgcctttgt-3’ for H233Y. The same PCR-based protocol was utilized to replace the coding sequence for amino acids 44–82 for a glycine, or a glycine and a cysteine in the deletions del-G and del-GC, respectively. Primer pairs for del-G and del-GC mutants were 5’-gccggacggaggaaaaaatatcggcaacgctaatgccg-3’, 5’-ttttttcctccgtccggcccgacgctgaagaaaaagctgg-3’ and 5’-ccggacggatgtggaaaaaatatcggcaacgctaatgccg-3’, 5’-tttttccacatccgtccggcccgacgctgaagaaaaagctg-3’, respectively.

ArnD point mutants D9N and H233Y were expressed and purified as described above for the wild-type protein. For the ArnD deletions del-G and del-GC, the plasmids were introduced in chemically competent E. coli BL21 (DE3) cells and single colonies from an overnight cultured plate used to inoculate a starter culture in LB media supplemented with 50 μg/mL kanamycin, which was then incubated at 37°C overnight. The starter culture was used to inoculate 1 L of LB media supplemented with 50 μg/mL kanamycin. The cells were grown at 37°C until the OD₆₀₀ = 0.6. The cultures were then cooled down on ice before adding IPTG (final concentration 0.1 mM) to induce protein expression. The cultures were then incubated overnight at room temperature with shaking. Cells were harvested by centrifugation at 4000 rpm (JLA 8.1000 rotor) for 30 minutes at 4°C. Cell pellets were resuspended in 50 mL of lysis buffer containing 25 mM Tris-HCl pH 8.0, 300 mM NaCl, 10% (v/v) glycerol, 5 mM β-mercaptoethanol (BME), lysozyme 0.5 mg/mL and 1 EDTA-free protease inhibitor cocktail tablet (Roche). The cells were lysed by 3 to 4 passes through an Avestin Emulsiflex C3 unit followed by removal of cell debris by centrifugation at 15000 rpm (Sorvall SS-34 rotor) for 30 minutes at 4°C. The supernatant was ultracentrifuged at 50000 rpm (Beckman fixed-angle Titanium rotor Type 70) for 2 hours at 4°C. An SDS-PAGE of the supernatant and pellet fractions revealed that del-G and del-CG proteins fractionated with the soluble fraction (supernatant) rather than the membranes (pellet). The pellet was then discarded, and the supernatant loaded on a 5 mL HisTrap FF Ni-NTA column (Cytiva) equilibrated in 25 mM Tris-Base pH 8.0, 300 mM NaCl, 10% (v/v) glycerol, 5 mM BME. The column was washed with 5 column volumes of the above buffer supplemented with 5 mM imidazole and the elution was achieved by increasing the imidazole concentration to 300 mM during a five-column volume linear gradient. Fractions containing ArnD, as evaluated by SDS-PAGE, were pooled, loaded in a HiLoad 26/60 Superdex 200 (Amersham Pharmacia Biotech) column equilibrated in 25 mM Tris-Base pH 8.0, 150 mM NaCl, 10% (v/v) glycerol, 5 mM BME, 1 mM EDTA and eluted in the same buffer. Fractions containing ArnD as detected by SDS-PAGE were pooled and concentrated to 18 mg/mL for del-G and 15 mg/mL for del-GC using a 10k MWCO centrifugal filter unit. Protein stocks were snap-frozen in liquid nitrogen and kept at −80°C for further use.

ArnD crystallization, x-ray data collection, and structure determination

Crystals of ArnD were obtained by mixing 2 μL of a stock solution of ArnD (6 mg/mL) in 25mM Tris-HCl, pH8.0, 150mM NaCl, 10% glycerol, 5mM BME, 0.0174% DDM, with 2 μL of a precipitant solution containing 0.1M Na citrate pH 6.2, 10% isopropanol and 17% PEG4000 and allowing vapor diffusion between the protein droplet and a 1mL reservoir of precipitant in a hanging drop set up incubated at 16 °C. Small crystals formed after 3–6 days, and they were harvested, and flash frozen in a stream of nitrogen (100 °K) prior to x-ray data collection. A dataset was collected at beam line 8.2.2 of the Advanced Light Source of the Lawrence Berkeley National Laboratory, which showed anisotropic diffraction to 2.6Å resolution. Data were indexed, integrated and scaled using HKL2000²². Data collection statistics are shown in SI, Table S1.

The crystals belonged to space group I222 with unit cell dimensions a = 82.55, b = 85.65, c = 111.91 Å suggesting one ArnD molecule per asymmetric unit. The structure was determined by molecular replacement using Phaser as implemented in the Phenix program suite^{23, 24}. The search model was produced by Dr. Kathryn Tunyasuvunakool, senior scientist at DeepMind, using AlphaFold2. The model was trimmed by removing the regions with residues with higher error estimates which included residues 44–84, 184–186, 201–208, 225–226 and 283–288. Rotation and translation searches using the trimmed model resulted in a single unambiguous solution. Rigid body refinement of the solution yielded a crystallographic R_free of 0.331 and inspection of the electron density maps revealed good density for most of the missing regions in the model confirming the correctness of the molecular replacement solution. The model was improved through alternating cycles of positional and B factor refinement in Phenix, and manual rebuilding using Coot^{25, 26}. Refinement statistics for the final model are shown in SI Table S1.

Enzymatic synthesis of undecaprenyl-phospho-L-[¹⁴C]Ara4FN (C55P-[¹⁴C]Ara4FN)

Since the ArnD substrate, C55P-Ara4FN, is not commercially available, its synthesis was carried out by sequential enzymatic conversion of UDP-glucuronic acid into C55P-Ara4FN using purified ArnA, ArnB and ArnC. Full-length E. coli ArnA and ArnB were expressed and purified as described^{27, 28}. Salmonella ArnC was cloned, expressed in E. coli, and purified using the same protocol described above for Salmonella ArnD. To obtain a radioactively labeled substrate, conversion of UDP-[¹⁴C]-glucuronic acid to UDP-L-[¹⁴C]-Ara4N was achieved preparing a 50 μL reaction mixture containing 50 mM Tris-Base pH 8.0, 10 mM BME, 6 mM NAD⁺ and 12 μM UDP-[¹⁴C]-glucuronic acid (Perkin Elmer, uniformly labeled with [¹⁴C] in the glucuronic acid, specific activity: 313 mCi/mmol), and mixing it with an equal volume of a solution containing 100 mM Glutamic acid pH 7.0, and the enzymes ArnA 0.3 mg/mL and ArnB 0.5 mg/mL. After a 3 hour incubation at room temperature, the reaction mixture was mixed with an equal volume of a solution containing 2 mM N₁₀-formyl-tetrahydrofolate, freshly prepared from N-5-formyltetrahydrofolate (Sigma) as previously described²⁹, to obtain the formylated amino sugar nucleotide UDP-L-[¹⁴C]Ara4FN. The reaction progress was followed by spotting 1 μL of each reaction mixture on a polyethylenimine (PEI) cellulose plate developed in a solvent system containing 0.25 M acetic acid and 0.4 M LiCl²⁹ and visualizing radioactive spots. The final mixture was filtered through a 30k MWCO centrifugal unit to remove the enzymes. To transfer the formylated sugar [¹⁴C]Ara4FN to a lipid acceptor, the filtered reaction mixture was supplemented with ArnC reaction components so that the final concentrations were: 50 mM HEPES pH 7.5, 0.2 mM MnCl₂, 0.2% (w/v) DDM, purified ArnC 0.25 mg/mL and 0.1 mM undecaprenyl-phosphate (Larodan). The reaction was incubated at room temperature for 1 hour which results in complete transfer of [¹⁴C]Ara4FN to the lipid to yield C55P-[¹⁴C]Ara4FN, as evaluated by spotting 1 μL of the reaction on a silica plate developed in a solvent system consisting of chloroform/methanol/water/ammonium hydroxide in a ratio of 35/40/3.6/0.4 and visualizing radioactive spots.

ArnD in vitro reconstituted activity assays.

ArnD activity measurements were typically carried out in 10 μL reaction consisting of 5 μL the ArnC reaction described above containing C55P-[¹⁴C]Ara4FN and 5 μL of ArnD 0.5 mg/mL supplemented with the indicated metals or EDTA. The mixtures were incubated at room temperature for the indicated times and the reaction progress evaluated by spotting 1 μL of the mixture on a silica plate developed in chloroform/methanol/water/ammonium hydroxide in a ratio of 35/40/3.6/0.4. Radioactive spots were visualized by exposing the TLC plates to phosphorimager screens and reading them on a Typhoon scanner. The reactions comparing activity of wild-type ArnD and mutants D9N and H233Y were carried out as described above but the final protein concentration was 0.1 mg/mL.

Results

ArnD is responsible for deformylation of undecaprenyl-phospho-4formamido-4deoxy-L-arabinose in E. coli.

Polymyxin resistance in E. coli, S. Thyphimurium, Pseudomonas aeruginosa and Klebsiella pneumoniae is primarily associated with lipid A modification with Ara4N as catalyzed by the integral membrane protein ArnT, which uses C55P-Ara4N as the donor substrate. However, in the pathway for biosynthesis of the ArnT substrate, ArnC efficiently transfers the formylated amino sugar 4-deoxy-4-formamido-L-Arabinose (Ara4FN) but not the unformylated 4-deoxy-4-amino-L-Arabinose from the UDP carrier to the undecaprenyl-phosphate intermediate to yield C55P-Ara4FN (SI Fig. S2 and S3). As the formylated amino sugar is not detected in lipid A a deformylation activity is required in the pathway. Based on weak sequence similarity to polysaccharide deacetylases, ArnD was proposed to catalyze the deformylation step. To test the involvement of ArnC and ArnD in the biosynthesis of C55P-sugar intermediates that lead to lipid A modification with Ara4N and polymyxin resistance in E. coli, phospholipids of polymyxin sensitive and resistant strains were analyzed by ESI/MS. As shown in Fig. 1A, only C55P [M-H]⁻ion at m/z 845.67 is detectable in the polymyxin sensitive E. coli reference strain W3110, consistent with non-expression of the arnBCADTEF operon under rich media growth conditions. Conversely, the polymyxin resistant E. coli pmrA^c strain, carrying a pmrA mutation that results in constitutive expression of the Arn operon, displays both C55P and C55P-Ara4N [M-H]⁻ ions at m/z 845.67 and 976.71 respectively (Fig. 1B). The spectrum also reveals the presence of a triply charged species (m/z 1007.2, labeled with an asterisk) that corresponds to CydH, a small subunit of the cytochrome bd-I oxidase often observed in Blight-Dyer extracts of E. coli^{30, 31} as well as its methionine oxidized derivative (m/z 1022.5). However, no C55P-Ara4N is observed, indicating that the formylated species is an intermediate in the pathway that does not accumulate. Deletion of the arnC gene in E. coli pmrA^c (pmrA^c ΔarnC) results in loss of C55P-Ara4N consistent with the role of ArnC in the transfer of Ara4FN from UDP to C55P (Fig. 1C). On the other hand, deletion of arnD (pmrA^c ΔarnD) results in accumulation of the formylated intermediate C55P-Ara4FN shown in the spectrum at m/z 1004.71 (Fig. 1D) suggesting that ArnD is responsible for deformylating the ArnC product in E. coli cells.

Figure 1: Negative ion ESI/MS spectra of the undecaprenyl-phosphate sugars in E. coli strains.

Lipids extracted from E. coli strains were subjected to negative ion ESI/MS. Only regions of the spectra showing the [M-H]⁻ ions of undecaprenyl-phosphate and undecaprenyl-phosphate sugars are shown. (A) Wild-type, polymyxin-sensitive, E. coli W3110. (B) polymyxin resistant E. coli W3110 pmrA^c. (C) E. coli W3110 pmrA^c ΔarnC. (D) E. coli W3110 pmrA^c ΔarnD. The exact masses for C55P, C55P-Ara4N and C55P-Ara4FN are 846.665, 977.724 and 1005.719, respectively. The asterisk (*) denotes chloroform soluble CydH.

Crystal structure of ArnD from Salmonella Typhimurium

Multiple constructs of ArnD from E. coli, S. Typhimurium, P. aeruginosa, and Yersinia pseudotuberculosis, with N- and C-terminal tags were expressed in E. coli and screened for purification and crystallization. Despite not having any predicted transmembrane regions, the proteins partitioned with the membrane fractions after cell lysis. Therefore, they were extracted and purified in the presence of detergents and screened for crystallization. Full length ArnD from Salmonella Typhimurium (stArnD) with a C-terminal hexahistidine tag yielded crystals amenable to x-ray diffraction analysis. The crystals belonged to space group I222 with one molecule per asymmetric unit. The crystal structure was determined by molecular replacement using an AlphaFold2 model and refined to 2.6 Å resolution as described in Materials and Methods. The final model comprises residues 1–297 with two breaks in the chain between residues 204–208, and 49–73 as no clear electron density was visible for these segments, presumably due to conformational flexibility (see below). The somewhat high crystallographic R factors of the refined model (R_work=0.225, R_free=0.288) likely reflect the missing residues in the model that has otherwise good stereochemistry and quality parameters as evaluated by the program Molprobity^{32, 33}. Complete X-ray data collection and refinement statistics are shown in SI Table S1.

The crystal structure of ArnD (Fig. 2A–B) reveals a NodB homology domain fold with a distorted (β/α)₇ barrel topology similar to the carbohydrate esterase family 4 (CE4) in the Carbohydrate Active Enzymes (CAZy) database³⁴. The peptidoglycan deacetylase PgdA is a well-characterized representative of the CE4 family and its structure adopts the characteristic distorted (β/α)₇ barrel in which one side of the barrel is formed by irregular secondary structure elements and the active site metal bound at the center of the barrel³⁵ (Fig 2C–D). ArnD superimposed on the PgdA structure with an RMS deviation of 1.9Å over 175 α-carbons despite only having 17% sequence identity (SI Mov. S1). However, whereas the overall fold is shared with polysaccharide esterases such as PgdA, analysis of the structure of stArnD and its sequence conservation (SI Fig. S4) indicate several divergences.

Figure 2: Crystal structure of Salmonella Typhimurium ArnD.

Top (A) and side (B) views of a cartoon representation of stArnD. Strands, helices, and coils in the NodB fold are colored light pink, pale cyan and light orange, respectively. The partially modeled 44 amino acid insertion is colored red and a C-terminal β-hairpin extension is colored blue. (C-D) Cartoon representation of the PgdA structure colored and oriented as in A and B. The bound metal ion is shown in red.

From a global fold standpoint, ArnD differs from the catalytic domain of PgdA in that a C-terminal extension forms a β-hairpin (blue in Fig. 2A–B) that packs against the distorted side of the β-barrel. Also notable is a 44 amino acid insertion protruding from the NodB fold between amino acids 41 and 85 (red in Fig 2A–B). This is only partially modeled in the crystallographic model due to conformational flexibility (see below). Previous analysis of the PgdA structure identified five conserved sequence motifs (MT1–5) associated with the polysaccharide deacetylase family³⁵. These motifs are partially or completely divergent in ArnD proteins (SI Mov. S2). In PgdA, motif 1, TFDD, contains D275 that binds acetate and plays a role in catalysis, and D276 that binds Zn²⁺; while motif 2, HS/TxxHP, contributes H326 and H330 to the Zn² binding site (Fig. 3A). In ArnD, motifs 1 and 2 have consensus sequences K/RxDVD and HxxD/NH respectively (Fig. 3B–C). Despite the partial divergence in the sequences, stArnD D9 and D11 in motif 1 as well as H106 and H110 in motif 2 conserve the three-dimensional arrangement observed in PgdA. This suggests they play similar roles in metal binding and catalysis even though our crystal structure did not display interpretable electron density at the site of metal binding. Motifs 3 and 4 in PgdA line the edges of the substrate binding groove and are defined as RPPY and DW respectively (Fig. 3D). In ArnD, motif 4 is instead composed of acidic resides (DE), and motif 3 is dominated by small side chain amino acids and a conserved C-terminal tryptophan with a consensus sequence AAPGW (AAAGW in stArnD) while the side chain of R364 in PgdA is instead contributed by the conserved R7 from motif 1 in ArnD (Fig. 3E–F). These divergences likely reflect the glycolipid nature of the ArnD substrate, C55P-Ara4FN, instead of glycan substrates in polysaccharide deacetylases such as PgdA. The last sequence motif in PgdA (MT5) is composed by the side chains of W385 and L415 that form a hydrophobic pocket to accommodate the methyl of acetyl groups in the substrates, and H417 that interacts with an acetyl oxygen and plays a role in catalysis (Fig. 3D). In ArnD, the hydrophobic pocket residues are not conserved (V195 and T231 in stArnD). Instead, the pocket is occupied by the side chain of W152 from motif 3 (Fig. 3E–F) resulting in a shallower binding site consistent with the formylated (rather than acetylated) substrate C55P-Ara4FN. The putative catalytic histidine is conserved (H233) and is surrounded by two conserved glutamates (E235 and E237, Fig. 3E). We thus propose that for ArnD proteins, motif 5 has a consensus sequence HxExEG (Figure 3F).

Figure 3: Conserved sequence motifs in ArnD vs Polysaccharide Deacetylases.

(A and D) Five motifs (MT1–5) previously described for the peptidoglycan deacetylase PgdA as a representative of polysaccharide deacetylases are shown in the active site of PgdA colored in light tints of pink (MT1), yellow (MT2), green (MT3), orange (MT4) and cyan (MT5). Bound Zn²⁺ ion and acetate are shown in red and black, respectively. (B and E) The corresponding but divergent motifs in ArnD are shown in bright hues of the same colors and a putative metal ion is modeled as a semitransparent sphere for reference. (C and F) Logos representations of sequence conservation in the five ArnD motifs, highlighted in boxes matching the color scheme in B-E.

ArnD membrane association

As ArnD partitions with the membrane fractions during purification despite the lack of predicted transmembrane regions, we sought to identify structural features that may mediate its membrane association. One salient characteristic of ArnD proteins is the 44 amino acid insertion that protrudes from the NodB fold (red in Fig. 2A–B). As this stretch of polypeptide could only be partially modeled in the crystal structure, AlphaFold2 predictions of the structure were utilized to develop a complete model. AlphaFold2 produced models with NodB domain architectures essentially identical to the crystal structure (SI Mov. S3, RMSD = 0.35Å). However, the 44 amino acid insertion resulted in alpha helical models that clustered in three different conformations with respect to the NodB domain (Fig. 4A). This apparent conformational flexibility is consistent with the lack of clear electron density in the crystal structure. AlphaFold2 models of the three representative conformation were submitted to the Positioning of Proteins in Membranes (PPM) server that utilizes the protein structure and amino acid composition to estimate the free energy of protein transfer from water to a membrane environment to predict membrane association^{36, 37}. As shown in Fig. 4B, all three ArnD models resulted in a peripheral association with the membrane, in all cases mediated by the flexible 44 amino acid insertion of ArnD.

Figure 4: ArnD membrane association models.

(A) stArnD with the NodB fold colored as in Fig. 2 with AlphaFold2-generated models for the 44 amino acid insertion. Three representative conformations for the insertion are shown colored in orange, yellow and green. (B) Membrane association models calculated using the PPM server for the three representative AlphaFold2 conformations shown in panel A. (C) Representation of the solvent accessible surface for the ArnD AlphaFold2 model with the membrane-interacting “finger” in the conformation depicted at the bottom of panel B, shown from the membrane plane. The surface is colored by the non-polar to polar residue ratio in a scale from 0.6 (polar) to 2.3 (non-polar). In all panels, a metal ion in the putative active site is shown as a red sphere.

The predictions obtained with the PPM server are consistent with those obtained with two additional predictive methods: the Membrane Optimal Docking Area (MODA) server based on the statistical preference of protein surface atoms for interaction with the membrane³⁸; and the DREAMM server that utilizes a machine learning approach trained on a set of 54 peripheral membrane proteins with amino acids experimentally determined to penetrate the membrane³⁹. Both algorithms also identified ArnD as a peripheral membrane protein and the residues predicted to mediate the interaction with the membrane mapped to the 44 amino acid insertion in all three structural models (SI Fig. S5).

To experimentally test these predictions, we engineered two variants of stArnD where amino acids 44–82 were replaced by a glycine (del-G) or a glycine-cysteine (del-GC) linker, which eliminates most of the 44 amino acid insertion in stArnD. Expression of both variants resulted in proteins that no longer fractionated with the membrane fractions and could be purified in the absence of detergents, yielding single peaks in size exclusion chromatography without signs of aggregation (SI Fig. S6). We thus conclude that the 44 amino acid insertion in ArnD proteins constitutes a membrane interaction “finger” that recruits the enzyme to act on its membrane soluble C55P-Ara4FN substrate. Consistent with this idea, the putative membrane interacting finger is located close to the active site (denoted by the modeled metal atom shown in red in Fig. 4B). Mapping the ratio of non-polar to polar residues on the solvent accessible surface of stArnD, as calculated with the protein-sol tool⁴⁰, reveals a hydrophobic “ramp” that may allow access of the C55P-Ara4FN substrate from the membrane to the metal binding active site in ArnD (Fig. 4C).

Reconstitution of ArnD activity in vitro

The analysis of C55P compounds derived from E. coli strains described above (Fig. 1), suggested that ArnD is responsible for deformylating the ArnC product C55P-Ara4FN in vivo. To unambiguously assign the ArnD activity, we developed an in vitro assay to use with purified recombinant stArnD. As the C55P-AraFN substrate is not commercially available, purified ArnA, ArnB and ArnC were used to convert UDP-[¹⁴C]Glucuronic acid into C55P-[¹⁴C]AraFN which results in a fast migrating species in silica thin-layer chromatography (TLC) as previously reported¹⁶ (Fig. 5A, lane 1). This product is then converted to a slower migrating species in the presence of purified ArnD, consistent with deformylation of C55P-[¹⁴C]Ara4FN to the more polar C55P-[¹⁴C]Ara4N (Fig. 5A, lane 2) indicating the reconstitution of ArnD activity in vitro.

Figure 5: ArnD is a metal-dependent lipid-phospho-L-Ara4FN deformylase.

(A) Thin layer chromatography separation of ArnD incubations with undecaprenyl-phospho-L-Ara4FN. UDP-Ara4FN was converted to C55P-Ara4FN by ArnC, and the product was incubated with purified ArnD for 10 minutes under the conditions indicated in each lane. (B) Thin layer chromatography separation of neryl-phosphate reactions with ArnC and ArnD. UDP-Ara4FN (lane 1) was incubated with neryl-phosphate plus (lane 2) or minus (lane 3) ArnC. The formylated amino-sugar is efficiently transferred by ArnC to the lipid to yield neryl-P-Ara4FN (C10P-Ara4FN). Lanes 4–8: The product of the ArnC reaction was incubated with purified ArnD for 10 minutes under the conditions indicated in each lane.

As the crystal structure indicated the conservation of a metal binding motif in ArnD, we next investigated the metal dependency of the in vitro activity. In situ synthesis of C55P-[¹⁴C]Ara4FN requires MnCl₂ for the activity of ArnC. Therefore, the ArnD reaction shown in lane 2 of Fig. 5A was carried out in the presence of 100 μM Mn²⁺. Supplementation of the reaction with 200 or 500 μM EDTA severely impairs activity, while further addition of an excess over EDTA of 20 μM Mn²⁺ recovers activity indicating that ArnD can utilize Mn²⁺ as a metal to catalyze the deformylation of C55P-[¹⁴C]Ara4FN (Fig. 5A, lanes 3–5). Addition of 20 μM Co²⁺ in a background of 100 μM Mn²⁺ strongly stimulates ArnD activity, whereas 20 μM Zn²⁺ does not increase the activity above the levels observed for Mn²⁺ (Fig. 5A, lanes 2, 6–7). Furthermore, in a background 100 μM Mn²⁺, addition of 200 μM Co²⁺ stimulates ArnD activity while 200μM Zn²⁺ inhibited the reaction (Fig. 5A, lanes 8–9). We conclude that the deformylase activity of ArnD is metal dependent consistent with the conserved features of the active site observed in the crystal structure.

We next tested the ability of ArnD to deformylate neryl-phospho-L-Ara4FN (C10P-Ara4FN), a shorter, diprenyl analog of C55P-Ara4FN. In contrast to undecaprenyl-phosphate that is difficult and expensive to obtain, neryl-phosphate can be easily synthesized from inexpensive nerol (SI supplementary methods). ArnC was able to convert UDP-[¹⁴C]Ara4FN into a lipid derivative in the presence of neryl-phosphate and Mn²⁺ indicating the synthesis of C10P-Ara4FN (Fig. 5B lanes 1–3). This product was efficiently deformylated (10 minutes) in the background of 1mM Mn²⁺. Sub-stoichiometric addition of EDTA to preferentially chelate any Co²⁺ that may be present in the reaction over the background Mn²⁺ still supported the deformylation catalyzed by ArnD (Fig. 5B lanes 4–6) further indicating that ArnD can utilize Mn²⁺ to deformylate its substrates. Similar to the results obtained with C55P-Ara4FN, the reaction was strongly stimulated by addition of 250 μM Co²⁺ but was inhibited by 250 μM Zn²⁺ (Fig. 5B lanes 7–8). Taken together, these results indicate that ArnD is responsible for deformylating the ArnC product C55P-Ara4FN and the reaction can be reconstituted in vitro in the presence of Mn²⁺ or, preferentially, Co²⁺. Furthermore, similar levels of activity are observed for the 10-carbon isoprenoid C10P-Ara4FN, which may facilitate future quantitative evaluations of ArnD activity and inhibition.

The crystal structure revealed that, in addition to the metal binding residues, an aspartate-histidine pair proposed to play roles in catalysis in polysaccharide deacetylases are conserved in the putative active site of stArnD. They are represented by D9 in motif 1 and H233 in motif 5 of stArnD. To test their potential involvement in catalysis, the deformylation activity of wild-type, D9N and H233Y mutants of ArnD was tested using C10P-[¹⁴C]Ara4FN as a substrate. As shown in Fig. 6A, both mutants were completely inactive, with no detectable activity even after a 16-hour incubation. We thus suggest that ArnD deformylates its substrates with a reaction mechanism analogous to that proposed for the peptidoglycan deacetylase PgdA³⁵. In such mechanism (Figure 6B), a water molecule is coordinated and polarized by the strong Lewis acid metal, while D9 acts as a general base abstracting its proton. The resulting hydroxide attacks the electrophilic carbonyl carbon of the formyl group in the substrate leading to a tetrahedral intermediate where the oxyanion is stabilized by the metal. Acting as a general acid, H233 donates a proton to the nitrogen in the tetrahedral intermediate, which resolves with release of the amino sugar-lipid and formic acid products.

Figure 6: Proposed ArnD mechanism.

(A) Thin layer chromatography separations of reactions containing C10P-[¹⁴C]Ara4FN and wild type (WT) or mutant (D9N or H233Y) ArnD. Samples were taken at the indicated time points of the reaction. (B) Proposed reaction mechanism based on similarity to polysaccharide deacetylases (adapted from reference³⁵, copyright (2005) National Academy of Sciences, U.S.A.). A coordinated divalent metal is denoted as Me²⁺.

Discussion

Modification of lipid A with Ara4N is one of the main mechanisms of bacterial resistance to innate immunity cationic antimicrobial peptides and the polymyxin family of antibiotics, of which colistin is the most clinically relevant. The proteins encoded in the pmrE(ugd) and arnBCADTEF loci constitute a biosynthetic pathway that converts the common metabolite UDP-Glucose into UDP-Ara4N and then into C55P-Ara4N, which serves as the amino sugar donor for lipid A modification. Interestingly, the N-terminal domain of ArnA catalyzes the N-formylation of UDP-Ara4N and the next enzyme in the pathway, ArnC, specifically utilizes the formylated intermediate for transfer to the C55P lipid carrier to yield C55P-Ara4FN. A deformylase activity is then required. We show unequivocally that ArnD is responsible for C55P-Ara4FN deformylation. Deletion of the arnD gene in pmrA^c mutants of E. coli that constitutively express the arnBCADTEF operon leads to accumulation of C55P-Ara4FN. This intermediate was efficiently deformylated in vitro by purified ArnD to yield the final amino-sugar donor C55P-Ara4N.

The crystal structure of stArnD revealed an overall NodB fold and conservation of a metal binding active site similar to the carbohydrate esterase family 4 (CE4) of carbohydrate active enzymes. This family is composed of metal-dependent deacetylases that act on polysaccharide (peptidoglycan, chitin, chitooligosaccharide and xylan) substrates. However, despite retaining the distorted (β/α)₇ barrel topology of the NodB fold, ArnD shares low overall sequence similarity to CE4 family members—17% sequence identity between stArnD and the representative CE4 enzyme PgdA. Furthermore, ArnD diverges in the sequence of five defining motifs as well as the presence of a 44 amino acid insertion and C-terminal β-hairpin extension. The insertion likely folds into an α-helical “finger” that mediates ArnD’s interaction with the membrane, and we show its deletion results in a soluble protein. Whereas some members of CE4 family such as Bacillus subtilis PdaC are membrane associated⁴¹, they do not contain the membrane interacting finger observed in ArnD and have separate membrane-association motifs. These divergent features suggest ArnD belongs in a separate carbohydrate esterase family, consistent with its distinct deformylase activity, membrane association motif and the glycolipid nature of its undecaprenyl-phosphate-L-Ara4FN substrate.

We show that stArnD efficient deformylation of C55P-Ara4FN is metal dependent with a preference for Co²⁺, although Mn²⁺ also supported the reaction. Promiscuity in metal binding dependence has been documented for multiple carbohydrate esterases and other enzymes^{35, 42–45}.The metal utilized in vivo is thus likely determined by the relative availability in the environment. Adak et al. have previously reported that chemically synthesized neryl-phosphate-L-Ara4FN (C10P-Ara4FN) could be slowly deformylated in an extended (20 h) incubation with Burkholderia cenocepacia ArnD (bcArnD) in the presence of 250 μM Zn²⁺¹⁷. This low level of activity could be due to bcArnD not being a bona fide deformylase, the diprenol C10P-Ara4FN not being the cognate substrate, or unfavorable in vitro reaction conditions. The experiments in Fig. 5 show that stArnD efficiently deformylates C10P-Ara4FN in the presence of Mn²⁺ or Co²⁺. However, stArnD deformylation of both C10P-Ar4FN and the cognate substrate C55P-Ara4FN is inhibited at high μM concentrations of Zn²⁺. This suggest that the relatively low activity of bcArnD was likely due to the presence of 250 μM Zn²⁺ in the in vitro reaction¹⁷.

The stArnD catalytic mechanism depended on the conserved D9 and H233 residues. Mutations D9N and H233Y—the most conservative mutations for aspartate and histidine residues respectively⁴⁶—completely abolished activity (Fig. 6). These results support a metal-assisted acid-base catalytic mechanism previously proposed for the deacetylase PgdA depicted in Fig. 6B. Fadouloglou et al. reported that two CE4 family polysaccharide deacetylases from B cereus and B. anthracis autocatalytically hydroxylate the Cα of a proline in the RPPY motif (MT3)⁴⁷. The modification, which resulted in approximately a 10-fold enhancement of activity over the non-hydroxylated enzyme, was proposed to aid in stabilization of the tetrahedral oxyanion intermediate, representing a form of active site maturation⁴⁷. Motif 3 is divergent in ArnD (AAAGW in stArnD, Fig 3E–F) and the proline hydroxylated in the bacilli polysaccharide deacetylases is neither conserved nor spatially contributed from elsewhere in the sequence, suggesting that this form of active site maturation is not observed in ArnD.

Colistin (polymyxin E) is a high priority antimicrobial in clinical use for treatment with multidrug resistant Gram-negative bacteria^48–50. However, colistin resistant isolates are increasingly reported^51–54. As arnD is essential for lipid A modification with Ara4N and polymyxin resistance¹¹, it is an attractive target for development of resistance inhibitors. There is ample precedent for inhibition of metal-dependent deacetylases as a clinically viable strategy^55–57. The structural and functional characterization of ArnD thus opens exciting opportunities for inhibitor development.

Abbreviations

Ara4N

4-deoxy-4-amino-L-arabinose

C55P-Ara4N

undecaprenyl-phospho-4-deoxy-4-amino-L-arabinose

C55P-Ara4FN

undecaprenyl-phospho-4-deoxy-4-formamido-L-arabinose

C10P-Ara4FN

neryl-phosphate-L- deoxy-4-formamido-L-arabinose

UDP-Ara4FN

Uridine-diphosphoL-4-deoxy-4-formamido-L-arabinose

pEtN

phosphoethanolamine

C55P

undecaprenyl-phosphate

DDM

n-Dodecyl-β-D-Maltopyranoside

IPTG

Isopropyl β-D-1-thiogalactopyranoside

BME

b-mercaptoethanol

Ni-NTA

nickel-nitrilotriacetic acid

MWCO

molecular weight cutoff

TLC

thin layer chromatography